Semantic segmentation of minimally invasive anti-reflux surgery video using U-NET Machine Learning
Abstract
Introduction: During anti-reflux surgery, there is a potential risk of Nervus Vagus (NV) injury. A solution is to create an AI tool that can detect the NV. and other anatomical structures in surgical videos. Addition of temporal features that might help in segmentation/detecting the actual nerve. Method: Five U-NET algorithms (different data inputs) are used as a for the training of five visible structures in the videoframes. These AI algorithms are trained a small clinical dataset for the actual goal and a larger automotive dataset for testing the functionality of the networks and testing the addition of Dense Optical Flow (temporal features). Results: The U-NETS can segment in all cases at least two structures with an IOU >0.5. Even three structures for the RGB only algorithm. No IOU above the 0.05 for the NV. Adding dense Optical Flow lowers in all cases the IOU. Although Dense Optical Flow only is still able to give spatial information. The confusion matrix shows similar findings. Conclusion: Training U-NETS and visualisation of the segmentation was possible on top of surgical video frames. Detecting the NV was not possible. The addition of temporal features by combining the RGB data does not improve the semantic segmentation.
Keywords
Not available
Specialisations
Similar theses
- Semantic segmentation of mammals in forest regions with U-NET Jhonatan Guilherme de Oliveira Cunha · Federal University of Technology – Paraná · 2024
- QUANTUM-INSPIRED NEURAL ARCHITECTURE SEARCH APPLIED TO SEMANTIC SEGMENTATION GUILHERME BALDO CARLOS · 2023
- Semantic segmentation of image sequences using a spatio-temporal U-Net Manuel Danner · TU Wien · 2020
- Transformers and contrastive semi-supervised learning for medical image segmentation Qianying Liu · The University of Glasgow · 2026
- Weakly supervised learning for medical image segmentation Hateef Alshewaier · Cardiff University · 2026
- Deep Learning for Medical Image Segmentation Binyan Hu · 2026
